基于机器视觉的煤矸石多工况识别研究
|
摘要
原煤入选前要进行预先排矸石操作,在多种工况下基于机器视觉对煤矸石进行识别。搭建图像采集装置采集煤块和块矸石图像,提取表面28个颜色和纹理特征参数,经过特征初步分析,将RGB空间特征作为冗余剔除。利用支持向量机作为分类器,并采用基于Relief算法权重的特征递归进一步筛选特征。将原煤表面状态分为外表面无煤泥且表面干燥、外表面无煤泥且表面湿润、外表面覆盖干煤泥、外表面覆盖湿煤泥4种类型。基于机器视觉对白芨沟矿的原煤进行识别试验,确定最优特征子集。在最优特征子集下进行多次随机取样识别试验,在4种不同工况下,5次随机实验的平均识别率大于等于94%,取得了满意的效果。
As gangue must be separated from raw coal preliminarily before coal preparation,we used machine vision to identify gangue in multi-condition. The image acquisition device was built to collect the images of coal and gangue,and 28 color and texture feature parameters were extracted. After the preliminary analysis of the feature,the RGB space features were removed as redundancy. The support vector machine was used as the classifier, and feature selection based on weights of Relief algorithm was used to further optimize the features. The surface conditions of the raw coal were divided into four types: slurryfree and dry surface,slurry-free and wet surface,surface covered with dry slurry,surface covered with dry slurry. The raw coal of Baijigou Mine was identified and the optimal feature subset was determined. With the optimal feature subset,another 5 experiments were carried out for validation. The average identification rate of 5 random experiments in each working condition was 94% at least,favorable results were achieved.
As gangue must be separated from raw coal preliminarily before coal preparation,we used machine vision to identify gangue in multi-condition. The image acquisition device was built to collect the images of coal and gangue,and 28 color and texture feature parameters were extracted. After the preliminary analysis of the feature,the RGB space features were removed as redundancy. The support vector machine was used as the classifier, and feature selection based on weights of Relief algorithm was used to further optimize the features. The surface conditions of the raw coal were divided into four types: slurryfree and dry surface,slurry-free and wet surface,surface covered with dry slurry,surface covered with dry slurry. The raw coal of Baijigou Mine was identified and the optimal feature subset was determined. With the optimal feature subset,another 5 experiments were carried out for validation. The average identification rate of 5 random experiments in each working condition was 94% at least,favorable results were achieved.
引文
[1]赵珊.中国煤炭资源现状及建议[J].广州化工,2014(15):52-53.
[2]郭秀军.煤矸石分选技术研究与应用[J].煤炭工程,2017(1):74-76.
[3]王嵘,王志良,马希荣.基于空间信息视觉注意模型的家庭机器人室内场景视频识别技术[J].机器人,2013,35(3):1239-1251.
[4]李红俊,韩冀皖.数字图像处理技术及其应用[J].计算机测量与控制,2002,10(9):620-622.
[5]王树文,闫成新,张天序,等.数学形态学在图像处理中的应用[J].计算机工程与应用,2004,40(32):89-92.
[6]马宪民,蒋勇.煤矸石二值图像的Roberts快速边缘检测法[J].仪器仪表学报,2005,26(S1):595-597.
[7]马宪民,田红,龚尚福.煤矸石在线识别与自动分选系统的研究[C].昆明:中国智能自动化会议,2001.
[8]马宪民,蒋勇.煤与矸石识别的数字图像处理方法探讨[J].煤矿机电,2004(5):9-11.
[9] Robnik-Sikonja M,Kononenko I. An adaptation of Relief forattribute estimation in regression[C]. Fourteenth InternationalConference on Machine Learning. Morgan Kaufmann PublishersInc. 1997:296-304.
[10]张泽琳,杨建国,王羽玲.基于MATLAB的煤粒图像识别系统及其密度和产率的预测[J].选煤技术,2011(1):53-56.
[11]张泽琳,杨建国,苏晓兰,等.基于图像分析的粗粒煤堆密度组成估计[J].中国矿业大学学报,2013,42(5):851-858.
[12]张泽琳.基于机器视觉的煤质快速分析方法研究[D].徐州:中国矿业大学,2014.
[13]王伟.基于图像处理和神经网络的煤与矸石分选系统的研究[D].青岛:山东科技大学,2013.
[14] Wu T,Xie K,Nie C,et al. An Adaption of Relief for RedundantFeature Elimination[M]. Advances in Neural Networks-ISNN2012. Springer Berlin Heidelberg,2012:73-81.
[2]郭秀军.煤矸石分选技术研究与应用[J].煤炭工程,2017(1):74-76.
[3]王嵘,王志良,马希荣.基于空间信息视觉注意模型的家庭机器人室内场景视频识别技术[J].机器人,2013,35(3):1239-1251.
[4]李红俊,韩冀皖.数字图像处理技术及其应用[J].计算机测量与控制,2002,10(9):620-622.
[5]王树文,闫成新,张天序,等.数学形态学在图像处理中的应用[J].计算机工程与应用,2004,40(32):89-92.
[6]马宪民,蒋勇.煤矸石二值图像的Roberts快速边缘检测法[J].仪器仪表学报,2005,26(S1):595-597.
[7]马宪民,田红,龚尚福.煤矸石在线识别与自动分选系统的研究[C].昆明:中国智能自动化会议,2001.
[8]马宪民,蒋勇.煤与矸石识别的数字图像处理方法探讨[J].煤矿机电,2004(5):9-11.
[9] Robnik-Sikonja M,Kononenko I. An adaptation of Relief forattribute estimation in regression[C]. Fourteenth InternationalConference on Machine Learning. Morgan Kaufmann PublishersInc. 1997:296-304.
[10]张泽琳,杨建国,王羽玲.基于MATLAB的煤粒图像识别系统及其密度和产率的预测[J].选煤技术,2011(1):53-56.
[11]张泽琳,杨建国,苏晓兰,等.基于图像分析的粗粒煤堆密度组成估计[J].中国矿业大学学报,2013,42(5):851-858.
[12]张泽琳.基于机器视觉的煤质快速分析方法研究[D].徐州:中国矿业大学,2014.
[13]王伟.基于图像处理和神经网络的煤与矸石分选系统的研究[D].青岛:山东科技大学,2013.
[14] Wu T,Xie K,Nie C,et al. An Adaption of Relief for RedundantFeature Elimination[M]. Advances in Neural Networks-ISNN2012. Springer Berlin Heidelberg,2012:73-81.